- 1 -
SPECIFICATION
{Title of Invention)
CONTROL DEVICE FOR PLANT AND CONTROL DEVICE FOR
THERMAL POWER PLANT
5 {Technical Field)
{OOOl)
The present invention relates to a control device
for a plant and more particularly to a control device
for a thermal power plant for generating power using
10 fossil fuel such as coal.
{Background Art)
(0002)
A control device for a plant processes measurement
signal data of the plant state quantity obtained from
15 the plant that is the control object, calculates a
control signal (an operation signal) given to the
control object, and controls the operation of the
control object.
{0003)
20 In the plant control device, an algorithm for
calculating the operation signal is implemented so that
the measurement signal data of the plant state quantity
satisfies the target value thereof.
(0004)
2 5 As a control algorithm used for plant control,
- 2 -
there is a PI (proportional integral) control algorithm
available. The PI control adds, to the product of the
deviation between the measurement signal data of the
plant state quantity and the target value thereof and
5 the proportional gain, the time integral value of the
deviation and derives the operation signal to be given
to the control object.
(0005)
The control algorithm using the PI control can
10 describe the input and output relation by a block chart,
so that the relation between cause and effect of input
and output is easy to understand and there are many
application results. However, when operating a plant
under conditions not assumed beforehand such as
15 operation state change and environmental change of the
plant, an operation of changing the control logic may
be necessary.
{OOOS}
On the other hand, as a control method applicable
20 to the change of the plant's operational status or
change of environment, there is a control method using
adaptable control that automatically corrects a control
algorithm and parameter values and a control method
using a learning algorithm.
2 5 (0007j
- 3 -
As a method for deriving the operation signal of
the plant control device using the learning algorithm,
a method is generally used where the data constructed
on the basis of the measurement data of the plant and
5 numerical calculation is used, those data is processed
statistically to construct a statistic model estimating
the plant properties, and the statistical model is made
to perform autonomous learning of an optimum control
logic.
10 ( 0008 }
It is known that the performance of the control
logic obtained by using the aforementioned method
depends on the estimation accuracy of the statistical
model. Namely, the learning algorithm learns the
15 control logic capable of obtaining a maximum control
effect for the statistical model, so that as the
property of the plant learned by the statistical model
approaches the actual plant property, the actual plant
control effect is increased. Therefore, in the
2 0 adaptable control technology using the learning
algorithm, the construction of a more highly precise
statistical model comes into a problem.
(0009)
Further, to this statistical model, the nonlinear
2 5 approximation method represented by a neural network
- 4 -
(NN) is applied in most cases, and this statistical
model learns a nonlinear curve expressing the plant
property using the plant measurement data. Further, in
recent years, due to the reason that the on-line
5 adjustment of the statistical model property becomes
easy at the time of plant application, the RBF (radial
basis function) network, which is one of the techniques
for NN, has been noticed and its application has been
being promoted.
10 {OOlO}
As one of the factors for determining the
estimation accuracy of the statistical model by the RBF
network, there is a radial parameter of the Gaussian
basis function for constituting the RBF network. The
15 RBF network arranges the Gaussian basis function given
by Formula (1) described later according to the
distribution of the model construction data used for
construction of the statistical model in the plant
property space, stacks up them, and thereby infers the
20 plant property.
{OOll}
Here, in Formula (I), i indicates an index of the
Gaussian basis function, bi a basis function value, x a
model input vector, ci a central coordinate vector of
2 5 the basis function, and ri a radial parameter.
(0012)
To obtain a statistical model with high estimation
accuracy, it is necessary to appropriately cover the
plant property space with the Gaussian basis function
5 and for that purpose, the radial parameter ri for
determining the spread of the Gaussian basis function
must be appropriately adjusted.
(0013)
(Formula 1)
(I)
(0014)
In view of the foregoing, as a technology of
15 improving the accuracy of the statistical model using
the RBF network, in Patent Literature 1, a technology
of adjusting the radial parameter of the Gaussian basis
function of the RBF network using the tunneling
algorithm which is one of the optimization algorithms
2 0 is described.
(0015)
Further, in Patent Literature 1, in consideration
of the roughness and fineness distribution of the data
used to learn the RBF network in the model property
25 space, a technology of optimally adjusting the radial
- 6 -
parameter is described.
{Citation List)
{Patent Literature)
{0016)
{Patent Literature 1)
Japanese Patent Laid-open No. 2005-115639
{Non Patent Literature}
{0017)
{Non Patent Literature I}
10 T. Eguchi, T. Sekiai, A. Yamada, S. Shimizu, M.
Fukai: "An adaptive Radius Adjusting Method for RBF
Networks Considering Data Densities and Its Application
to Plant Control Technology", in Proc. of ICCASSICE2009,
pp. 4188-4194 (2009)
15 {Summary of Invention)
{Problems to be Solved by the Invention)
{0018)
When the aforementioned technologies disclosed in
Patent Literature 1 and/or Non Patent Literature 1 are
2 0 applied to the plant control device, the radial
parameter is adjusted so that the Gaussian basis
function can appropriately cover the plant property
space, the estimation accuracy of the statistical model
can be improved.
(0019)
- I -
On the other hand, in the plant control, after
changing the operation conditions under control, it
takes several minutes to more than ten minutes before
the plant operation property is stabilized, so that
5 this time is generally set to a control cycle.
Therefore, the aforementioned online adjustment of the
statistical model (radial parameter adjustment and
plant property curve learning) desirably ends within
the control cycle.
10 {0020}
By the way, when applying the technology of Patent
Literature 1 to the plant control device, the
evaluation value for the solution candidate of the
radial parameter to be searched for by the optimization
15 algorithm must be calculated by the error evaluation
between the model construction data which is learning
data and the estimate result.
{0021}
The calculation cost required for the error
2 0 evaluation increases according to the number of pieces
of model construction data, so that when it is applied
to the plant control, as measurement data is
accumulated, the calculation time increases and there
are possibilities that the optimization calculation may
2 5 not be ended within the control cycle.
(00221
When the optimization calculation is not ended
within the control cycle due to the increase of the
calculation time, the estimation accuracy of the
5 statistical model lowers, so that the plant can be
hardly operated in a desirable state.
(0023)
Further, when applying the aforementioned
technology in Non Patent Literature 1 to the plant
10 control device, the aforementioned error evaluation is
unnecessary, though every time measurement data is
accumulated, the radial parameters of all the Gaussian
basis functions are adjusted in consideration of the
roughness and fineness between the model construction
15 data. Therefore, even in the adjustment of the radial
parameters by the aforementioned technology, similarly
to the case of the aforementioned technology of Patent
Literature 1, there is a possibility that the
calculation cost will be increased due to accumulation
2 0 of measurement data, so the optimization calculation
will not be ended within the control cycle; as a result,
the estimation accuracy of the statistical model will
lower, which makes it difficult to operate the plant in
a desirable state.
2 5 (00241
- 9 -
An object of the present invention is to provide,
when adjusting the statistical model online by the RBF
network using the plant measurement data, a control
device for a plant and a control device for a thermal
power plant that can end the optimization calculation
within the control cycle and that improves the
estimation accuracy of the statistical model.
{Means for Solving the Problems}
(0025)
The control device for a plant of the present
invention for taking in the measurement signal data,
which is the state quantity of the concerned plant,
from the plant and calculating the operation signal for
controlling the plant using the measurement signal data,
characterized in that, the control device is comprising
a measurement signal database for taking in and saving
the measurement signal data which is the state quantity
of the plant; a data pre-processing part for executing
a statistical process on the basis of a confidence
interval for the measurement data of the plant which
has been saved in the measurement signal database,
thereby generating model construction data; a model
constructive database for saving the model construction
data generated by the data pre-processing part; a
statistical model for simulating a control property of
- 10 -
the plant for estimating a value of the measurement
signal data which is the state quantity of the plant
when a control signal is given to the plant using the
model construction data saved in the model constructive
5 database; an operation learning part for learning a
generation method of model input equivalent to the
control signal given to the plant so that model output
equivalent to the measurement signal data accomplishes
its target value using the statistical model; a
10 learning information database for saving learning data
relating to learning restrictions and learning results
in the operation learning part; and a control signal
generation part for calculating the control signal
transmitted to the plant using the measurement signal
15 data of the measurement signal database and the
learning data of the learning information database,
wherein the control device is further comprising a
model adjusting part for adjusting a basis radial
parameter of the statistical model included in the
20 model construction data saved in the model constructive
database so that the statistical model generates the
model output using adjustment results of the basis
radial parameter by the model adjusting part and the
model output is output to the operation learning part.
I00261
- 11 -
The control device for a thermal power plant of the
present invention for taking in the measurement signal
data, which is the state quantity of the concerned
plant, from the thermal power plant including a boiler
5 and calculating the operation signal for controlling
the plant using the measurement signal data,
characterized in that, the control device is comprising
a measurement signal database for taking in and saving
the measurement signal data which is a state quantity
10 of the plant; a data pre-processing part for executing
a statistical process on the basis of a confidence
interval for the measurement data of the plant which
has been saved in the measurement signal database,
thereby generating model construction data; a model
15 constructive database for saving the model construction
data generated; a statistical model for simulating a
control property of the plant for estimating a
measurement signal data value which is the state
quantity of the plant when a control signal is given to
20 the plant using the model construction data saved in
the model constructive database; an operation learning
part for learning a generation method of model input
equivalent to the control signal given to the plant so
that model output equivalent to the measurement signal
2 5 data accomplishes its target value using the
- 12 -
statistical model; a learning information database for
saving learning data relating to learning restrictions
and learning results in the operation learning part;
and a control signal generation part for calculating
5 the control signal transmitted to the plant using the
measurement signal data of the measurement signal
database and the learning data of the learning
information database, wherein the control device is
further comprising a model adjusting part for adjusting
10 a basis radial parameter of the statistical model
included in the model construction data saved in the
model constructive database so that the statistical
model generates the model output using adjustment
results of the basis radial parameter by the model
15 adjusting part and the model output is output to the
operation learning part, and the measurement signal
data includes a signal of the state quantity expressing
at least one of concentrations of nitrogen oxide,
carbon monoxide, carbon dioxide, nitrogen, oxygen,
20 unburnt carbon in ash, and hydrogen sulfide, and the
control signal includes a signal expressing at least
one of an air flow rate fed to the boiler of the
thermal power plant, an aperture of an air dumper for
adjusting the air flow rate, a fuel flow rate supplied
2 5 to the boiler, and an exhaust gas recirculation flow
- 13 -
rate for recirculating exhaust gas discharged from the
boiler to the boiler.
{Advantageous Effects of Invention}
I00271
According to the present invention, when using the
measurement data of the plant and adjusting the
statistical model on line by the RBF network, a control
device for a plant and a control device for a thermal
power plant that can end the optimization calculation
10 within the control cycle and improves the estimation
accuracy of the statistical model can be implemented.
{Brief Description of Drawings}
(0028)
{Fig. 1) Fig. 1 is a block diagram showing the
15 constitution of the control device for the plant which
is a first embodiment of the present invention.
{Fig. 2) Fig. 2 is a flow chart showing a series of
control operation flow in the control device for the
plant of Embodiment 1.
{Fig. 3) Fig. 3 is a flow chart showing the
operation procedure of the pre-processing part in the
control device for the plant of Embodiment 1.
{Fig. 4 ) Fig. 4 is a schematic view showing the
concept of the model output reliability section
2 5 calculation of the pre-processing part in the control
- 14 -
device for the plant of Embodiment 1.
{Fig. 5) Fig. 5 is a diagram showing the aspect of
the data saved in the model constructive database in
the control device for the plant of Embodiment 1.
5 {Fig. 6) Fig. 6 is a schematic view of the RBF
network composing the statistical model in the control
device for the plant of Embodiment 1.
{Fig. 7) Fig. 7 is a flow chart showing the
operation procedure of the model adjusting part in the
10 control device for the plant of Embodiment 1.
{Fig. 8) Fig. 8 is a schematic view showing the
concept of the system for determining the Gaussian
basis function which is a radius adjustment object at
the operation time of the model adjusting part in the
15 control device for the plant of Embodiment 1.
{Fig. 9) Fig. 9 is a flow chart showing the
operation procedure of the radius adjustment algorithm
at the operation time of the model adjusting part in
the control device for the plant of Embodiment 1.
{Fig. 10) Fig. 10 is a schematic view showing the
concept of the contribution degree of the Gaussian
basis function which is a radius adjustment object at
the operation time of the model adjusting part in the
control device for the plant of Embodiment 1.
{Fig. 11) Fig. 11 is a diagram showing an example
- 15 -
of the screen displayed on the display device when
setting the execution conditions in the control device
for the plant of Embodiment 1.
{Fig. 12) Fig. 12 is a diagram showing an example
5 of the screen displayed on the display device when
selecting a learning result list to be displayed in the
control device for the plant of Embodiment 1.
{Fig. 13) Fig. 13 is a diagram showing an example
of the screen displayed on the display device when
10 displaying the statistical model property of the
learning results and the guidance conditions in the
control device for the plant of Embodiment 1.
{Fig. 141 Fig. 14 is a schematic block diagram
showing the constitution of the thermal power plant to
15 which the control device of Embodiment 2 is applied.
{Fig. 15) Fig. 15 is a schematic block diagram
showing the constitution of the air heater of the
thermal power plant to which the control device of
Embodiment 2 is applied.
2 0 {Description of Embodiments}
I0029)
The embodiments of a control device for a plant of
the present invention and a control device for a
thermal power plant will be explained below by
25 referring to the accompanying drawings.
{Embodiment I}
{0030}
A control device for a plant which is the first
embodiment of the present invention will be explained
5 by referring to Figs. 1 to 13.
{0031}
Fig. 1 is a system block diagram of the control
device for a plant which is the first embodiment of the
present invention. As shown in Fig. 1, in the control
10 device for a plant which is this embodiment, a plant
100 which is a control object is controlled by a
control device 200.
{0032}
The control device 200 controlling the plant 100 is
15 connected to a maintenance tool 910, so that an
operator of the plant 100, via external data input
equipment 900 and a display device (for example, a CRT
display) 920 which are connected to the maintenance
tool 910, can control the control device 200.
{0033}
The control device 200 is structured so as to
include, as operational equipment, a data preprocessing
part 300, a numerical calculation part 400,
a statistical model 500, a model adjusting part 600, a
25 control signal generation part 700, and an operation
learning part 800.
I00341
Further, the control device 200, as a database (DB),
is provided with a measurement signal database 210, a
5 model constructive database 220, a learning information
database 230, a control logic database 240, and a
control signal database 250.
I00351
Further, the control device 200, as an interface
10 with the outside, is provided with an outside input
interface 201 and an outside output interface 202.
to0361
In the control device 200, measurement signal data
1 that has measured the state quantity of the plant 100
15 taken in from the plant 100 is saved in the measurement
signal database 210 via the outside input interface 201.
(0037)
Further, a control signal 15 to be generated by the
control signal generation part 700 installed in the
2 0 control device 200 is structured so as to be saved in
the control signal database 250 installed in the
control device 200 and also be outputted as an
operation signal 16 for the plant 100 from the outside
output interface 202, for example, as an operation
2 5 signal 16 for controlling the air flow rate to be fed
to the plant 100.
{0038}
The data pre-processing part 300 installed in the
control device 200 converts measurement data 3 saved in
the measurement signal data 210 and numerical
calculation data 5 obtained by executing the numerical
calculation part 400 using a physical model that
simulates the behavior of the plant 100 to model
construction data 4 using a statistical process.
{0039}
The numerical calculation part 400 has a function
of calculating the operation property of the plant 100
by numerical calculation. Further, the numerical
calculation data 5 obtained by the numerical
calculation part 400 is a property value of the plant
100.
{0040}
The model construction data 4 is saved in the model
constructive database 220. Further, a part of the
measurement data 3 is input to the control signal
generation part 700 installed in the control device 200
{0041}
The model adjusting part 600 installed in the
control device 200 updates (adjusts the model) the
model parameter information included in model
- 19 -
construction data 7 taken in from the model
constructive database 220 and saves the model
construction data 8 after update in the model
constructive database 220.
5 (0042)
The operation learning part 800 installed in the
control device 200 generates learning data 12 and saves
it in the learning information database 230.
I00431
10 The statistical model 500 installed in the control
device 200 has a function of simulating the control
property of the plant 100 which becomes a control
object. Namely, the statistical model 500 gives the
operation signal 16 to the plant 100 and simulation-
15 calculates a function similar to obtaining the
measurement signal data 1 for the control results. For
this simulation calculation, the statistical model 500
uses model input 9 received from the operation learning
part 800 and model construction data 6 saved in the
2 0 model constructive database 220.
I00441
The model input 9 is equivalent to the operation
signal 16. From the model input 9 and the model
construction data 6, the statistical model 500, by the
25 RBF network which is one of the techniques of the
- 20 -
neural network composing the statistical model 500,
simulation-calculates the property change under the
control of the plant 100 and outputs model output 10.
(0045)
The model output 10 obtained by the statistical
model 500 is a predicted value of the measurement
signal data 1 of the plant 100. Further, both the model
input 9 and the model output 10 are not limited to one
kind of number and plural kinds can be provided.
10 Here, in the statistical model 500, as mentioned
above, it is a prior condition to use the RBF network,
though as the basis function, other than the Gaussian
function, a well-known function group (thin-platespline,
Inverse Multiquadrics, etc.) may be used. Even
15 in that case, the parameter for determining the spread
of the basis function is an adjustment object.
(0046)
The control signal generation part 700 installed in
the control device 200, using the learning information
20 data 13 output from the learning information database
230 and the control logic data 14 saved in the control
logic database 250, generates the control signal 15 so
that the measurement signal data 1 becomes a desirable
value.
(0047)
- 21 -
In the control logic database 250, the control
circuit for calculating the control logic data 14 and
the control parameter are saved. For the control
circuit for calculating the control logic data 14, as a
5 conventional technology, the well-known PI
(proportional integral) control can be used.
(0048)
The operation learning part 800, using the learning
information data 11 including the learning restrictions
10 and the learning parameter setting conditions which are
saved in the learning information database 230, learns
the operating method of the model input 9. The learning
data 12 which is learning results is saved in the
learning information database 230.
t0049)
As mentioned above, in the operation of the control
device 200, since a mechanism that the model parameter
information included in the model construction data 7
saved in the model constructive database 220 is
2 0 adjusted in the model adjusting part 600 is furnished,
an appropriate model parameter according to the
property of the model construction data 7 is provided,
so that the estimation accuracy of the plant property
in the statistical model 500 can be improved.
{0050)
- 22 -
Further, such a radial parameter adjustment, for
the measurement data added at the time of application
of the plant control device of the present invention,
only for the minimum Gaussian basis functions
5 distributed in the neighborhood, is executed, so that
the increase in the calculation cost due to
accumulation of the measurement data is avoided and
completion of the statistical model adjustment within
the control cycle can be expected.
10 {0051)
Further, the detailed functions of the data preprocessing
part 300, the statistical model 500, the
model adjusting part 600, and the operation learning
part 800 which are installed in the control device 200
15 will be described later.
(0052)
Further, in the learning data 12 saved in the
learning information database 230 from the operation
learning part 800, the information relating to the
20 model input before and after the operation and the
model output obtained as a result of the operation is
included.
{0053}
In the learning information database 230, the
25 learning data 12 corresponding to the present operation
- 23 -
conditions is selected and is input to the control
signal generation part 700 as the learning information
data 13.
{0054}
The operator of the plant 100, by use of the
external data input equipment 900 composed of a
keyboard 901 and a mouse 902, the maintenance tool 910
capable of transmitting and receiving data to and from
the control device 200, and the display device 920, can
10 access information saved in various databases installed
in the control device 200. The control device 200
includes an input part or an output part for
transferring the input-output data information 90 to
and from the maintenance tool 910.
15 to0551
Further, by use of these devices, the parameter
setting values, the learning restrictions, and the
setting information necessary for confirmation of the
obtained learning results which are used by the
2 0 numerical calculation part 400, the statistical model
500, the model adjusting part 600, and the operation
learning part 800 of the control device 200, can be
input.
{ 0056)
2 5 The maintenance tool 910 is composed of an outside
- 24 -
input interface 911, a data transmitting and receiving
part 912, and an outside output interface 913 and data
can be transmitted and received to and from the control
device 200 via the data transmitting and receiving part
5 912.
(00571
A maintenance tool input signal 91 generated by the
external data input equipment 900 is captured into the
maintenance tool 910 via the outside input interface
10 911. The data transmitting and receiving part 912 of
the maintenance tool 910, according to the information
of a maintenance tool input signal 92, obtains the
input-output data information 90 from the control
device 200.
(0058)
Further, the data transmitting and receiving part
912, according to the information of the maintenance
tool input signal 92, outputs the input-output data
information 90 including the parameter setting values,
20 the learning restrictions, and the setting information
necessary for visually recognizing the obtained
learning results which are used by the numerical
calculation part 400, the statistical model 500, the
model adjusting part 600, and the operation learning
25 part 800 of the control device 200.
(0059)
The data transmitting and receiving part 912
transmits a maintenance tool output signal 93 obtained
as a result of processing the input-output information
90 to the outside output interface 913. A maintenance
tool output signal 94 transmitted from the outside
output interface 913 is displayed on the display device
920.
(0060)
Further, in the aforementioned control device 200,
the measurement signal database 210, the model
constructive database 220, the learning information
database 230, the control logic database 240, and the
control signal database 250 are arranged inside the
control device 200, though all the devices or a part of
them can be arranged outside the control device 200.
(0061)
Further, the numerical calculation part 400 is
arranged inside the control device 200, though it can
be arranged outside the control device 200.
I00621
For example, it is possible to arrange the
numerical calculation part 400 and the model
constructive database 220 outside the control device
200 and transmit the numerical calculation data 5 to
- 26 -
the control device over the internet.
I00631
Fig. 2 is a flow chart showing the control
procedure of the control device for the plant of this
5 embodiment shown in Fig. 1.
(0064)
Fig. 2 is a flow chart showing the operation of the
control device 200 for the plant of this embodiment and
the flow chart is executed by a combination of Steps
10 1000, 1100, 1200, 1300, 1400, 1500, 1600, 1700, 1800,
1900, 2000, and 2100. Hereinafter, each Step will be
explained.
I00651
After the operation start of the control device 200,
15 Step 1000 for setting the execution conditions of the
control device 200 sets various parameter conditions
used when executing the data pre-processing part 300,
the statistical model 500, the model adjusting part 600,
and the operation learning part 800.
(0066)
Next, Step 1100 for obtaining the measurement
signal data measured by the plant, via the outside
input interface 201 installed in the control device 200,
obtains the measurement signal data 1 of the plant 100
2 5 and saves it in the measurement signal database 210 as
measurement data 3.
(0067)
Next, Step 1200 for executing the data preprocessing
operates the data pre-processing part 300 of
5 the control device 200, executes the statistical
process for the measurement data 3 saved in the
measurement signal database 210 and the numerical
calculation data 5 obtained by executing the numerical
calculation part 400 and converts it to the model
10 construction data 4. Further, the detailed function and
operation of the data pre-processing part 300 will be
described later.
(0068)
Next, Step 1300 saves the model construction data 4
15 converted by the pre-processing in the model
constructive database 220.
{0069}
Next, Step 1400 for adjusting the statistical model
operates the model adjusting part 600 of the control
20 device 200 and updates the radial parameter of the
Gaussian basis function arranged according to the model
construction data 7. Here, the model construction data
7 includes the model input value and model output value
in the model property space of each piece of data, the
2 5 radial parameter value of the Gaussian basis function
- 28 -
arranged on the data, and the weight parameter of the
RBF network composing the statistical model 500.
Further, the detailed function and operation of the
model adjusting part 600 will be described later.
5 {0070}
Next, Step 1500 for learning the statistical model
operates the statistical model 500 of the control
device 200 and learns the weight parameter of the RBF
network composing the statistical model 500. The
10 learning means can use a well-known scheme.
{0071)
Next, Step 1600 for learning the operation method
operates the operation learning part 800 of the control
device 200 and learns the operation method of the model
15 input 9 so that the model output estimation value 10
calculated by the statistical model 500 becomes lower
than the target condition preset via the display device
920. Further, the operation learning part 800 can use a
well-known scheme such as tempered learning and
2 0 learning information data 11 including the execution
conditions thereof is input from the learning
information database 230.
{0072)
Next, Step 1700 saves the learning result data 12
2 5 by the operation learning part 800 in the learning
- 29 -
information database 230.
I00731
Next, Step 1800 for generating a control signal
operates the control signal generation part 700 of the
5 control device 200 and using learning result data 13
and the control logic 14 which are saved in the
learning information database 230 and the control logic
database 240, generates the control signal 15. The
generated control signal 15 is used to control the
10 plant 100 and is saved in the control signal database
250.
(0074)
The next Step 1900 for determining control
execution is a branch. From the simulation results of
15 the plant operation based on the learning results, when
it can be seen that the control results become desired
results, the process goes to Step 2000 and in other
cases, the process goes to Step 2100.
(0075)
Next, Step 2000 for controlling the plant 100
outputs the generated control signal 15 as the control
signal 16 via the outside output interface 202
installed in the control device 200 and controls the
plant.
(0076)
- 30 -
Finally, Step 2100 for determining the end of a
series of process operation is a branch. When a signal
for ending the operation of the control device 200 of
the present invention via the external data input
5 equipment 900 is input, the process goes to the step of
ending the process and in other cases, the process
returns to Step 1100.
(0077)
By the above sequence, in the operation of the
10 control device 200 of this embodiment, on the basis of
the execution conditions set by the operator of the
plant 100, a series of processes of the data preprocessing,
model adjustment and learning, operation
method learning, control signal generation, and control
15 execution can be autonomously acquired and executed.
(0078)
Next, the detailed operation of the data preprocessing
part 300 of the control device 200 will be
explained by referring to the flow chart in Fig. 3 and
2 0 the schematic diagrams of Figs. 4, 5, and 6.
(0079)
The control device 200 of this embodiment
preprocesses the measurement data of the plant obtained
within the control cycle or the numerical calculation
25 data, thereby converts to the model input value and
- 31 -
model output value and generates model construction
data.
{0080)
The data pre-processing part 300 of this embodiment,
5 in consideration of the point that the time series
plant measurement data has a change, for the
measurement data including the change acquired within
the control cycle, executes the statistical process
based on the confidence interval and calculates three
10 kinds of model output values of maximum, average, and
minimum.
(0081)
Here, the confidence interval is an interval that
the probability that the true parameter (average,
15 fraction) value of the obtained measurement data
distribution enters a certain interval (L, U) is
guaranteed so as to be 1-8 or more, where L is referred
to as a lower confidence boundary, U an upper
confidence boundary, and 1-8 a confidence coefficient.
20 Namely, the model output values of the three kinds of
maximum, average, and minimum, according to the
confidence interval calculation for the measurement
data distribution, respectively correspond to the upper
confidence boundary, average, and lower confidence
2 5 boundary.
I00821
Fig. 3 is a flow chart showing the operation of the
data pre-processing part 300, showing the operation of
Step 1200 of the flow chart shown in Fig. 2 in detail.
5 The flow chart shown in Fig. 3 executes a combination
of Steps 1210, 1220, and 1230. Hereinafter, the
respective steps will be explained.
I00831
After the operation start of the data pre-
10 processing part 300, firstly, Step 1210 decides a bias
of the model input. Even when the plant 100 is in the
normal state, the measurement data has a change, so
that the model input conditions obtained by converting
the measurement data are also changed. Therefore, in
15 consideration of the bias for the average of the model
input obtained, the confidence interval for which
distribution is considered for the model output value
controlled within the range is calculated.
(00841
2 0 Fig. 4 is a schematic diagram (vertical axis: model
output value, horizontal axis: model input value) where
the distribution of the measurement data relating to
certain model output is plotted in the model input
space.
{0085}
- 33 -
In Fig. 4, x indicates an average of the model
input values obtained from the measurement data and Ax
indicates a bias. Here, the bias can be set to an
optional value, though generally, it is desirable to
5 use a standard deviation of the model input value of
the data.
I00861
Next, Step 1220 for calculating the model output
confidence interval, for the measurement data existing
10 in the model input range in consideration of the bias
shown in gray in Fig. 4, calculates each statistic
quantity of the maximum, average, and minimum values of
the data on the basis of the preset confidence
coefficient (1-8) .
15 (0087)
Next, Step 1230 for preparing the model
construction data saves the model construction data 4
including the model output values of the three kinds of
maximum, average, and minimum for the model input
2 0 conditions obtained at Step 1220 in the model
constructive database 220. Here, the saved aspect of
the model constructive database 220 and the
constitution of the statistical model will be explained
by referring to Figs. 5 and 6.
2 5 { 0088 1
- 34 -
In Fig. 5, the model construction data has
respectively peculiar data ID 221. Further, as shown in
Fig. 5, model input 222 of 1 case has model output 223
of 3 (maximum, average, minimum) x K (model output
5 count) kinds.
(00891
Fig. 6 is a schematic view showing the structure of
the RBF network composing the statistic model 500
installed in the control device 200 of this embodiment
10 and one case of the model construction data is
equivalent to one intermediate layer node. Therefore,
for one intermediate layer node, 3 x K Gaussian basis
functions are arranged. This indicates that in Fig. 5,
one piece of data (model input) has 3 x K radial
15 parameters 224 and weight parameters 225 of the RBF
network.
(0090)
Namely, on one network, three kinds of data
properties (maximum, average, minimum) and the plant
2 0 properties according to the model output count K can be
expressed and compared with the case that single
Gaussian basis function is arranged on the intermediate
layer node, an estimate with more highly accuracy is
enabled.
{009l)
- 35 -
The operation learning part 600 executes the
learning of the operation method in combination with
the statistical models 500 structured as mentioned
above (Step 1600 in Fig. 2). Concretely, for each model
5 output, using one kind of model construction data
selected from the three kinds of data properties
(maximum, average, minimum), the statistical models 500
are combined.
(0092)
10 Therefore, for the combinations of a maximum of kpower
of 3, the operation method can be learned. Due to
a reduction in the calculation cost of the model
adjustment of the present invention, the learning by
various combinations of the statistical models is
15 enabled.
(0093)
The operator of the plant, using the display device
and maintenance tool which will be described later,
among the learning results obtained by various
2 0 statistic model combinations, optionally selects the
learning result suitable for the control needs and can
execute the control of the plant.
{0094)
This ends the explanation relating to the operation
2 5 of the data pre-processing part 300 of the control
- 36 -
device 200 of the present invention.
(0095)
Next, the detailed operation of the model adjusting
part 600 in the control device 200 will be explained by
5 referring to the flow charts in Figs. 7 and 9 and the
schematic diagrams in Figs. 8, 11, and 12.
(0096)
Fig. 7 is a flow chart showing the algorithm
operation of model adjustment by the model adjusting
10 part 600 and is equivalent to Step 1400 for adjusting
the statistical model in the flow chart shown in Fig. 2.
(0097)
The flow chart shown in Fig. 7 executes a
combination of Steps 1410, 1420, 1430, 1440, and 1450.
15 {0098}
After start of the algorithm of the statistical
model adjustment, Step 1410 for calculating the
distance neighborhood for the additional data of each
basis function, for the model construction data newly
20 added at Step 1300 in Fig. 2, on the basis of the
distance from the existing Gaussian basis function,
calculates an index of the distance neighborhood. Here,
the distance between the Gaussian basis function and
the additional data is a Euclid distance between the
2 5 central coordinates of the basis function and the
- 37 -
coordinates of the additional data.
{0099)
Here, the calculation means of the distance
neighborhood will be explained by referring to the
5 schematic diagram in Fig. 8. Fig. 8 shows the central
coordinates of the Gaussian basis function in the twodimensional
model input space and the distribution of
the additional data. In the calculation of the distance
neighborhood, firstly, the coordinates of the
10 additional data are assumed as a branch point and the
model input space is divided.
{OlOO}
In the schematic diagram shown in Fig. 8, for the
two kinds of model input, the dashed lines shown on the
15 coordinates of the additional data are assumed as a
branch point and the region is divided into four parts.
Namely, assuming the degree of the model input as J,
the division number of the model input space is the Jpower
of 2. Assuming the Gaussian basis functions
20 included in each divided region as a group, the
distance neighborhood is calculated for each group.
{OlOl}
As shown in the schematic diagram in Fig. 8, the
distance neighborhood of the basis function that the
2 5 distance from the additional data in each group is
- 38 -
minimum is assumed as 1 and the distance neighborhood
is added one by one in the ascending order of distance
(the numerical value on the right shoulder of each
basis function indicates the distance neighborhood). As
5 the distance neighborhood becomes smaller, the
influence onto the basis functions due to data addition
increases and the necessity of the radial parameter
adjustment increases. Further, the distance
neighborhood is calculated for every divided region,
10 thus the Gaussian basis functions which are a radius
adjustment object are uniformly selected so as to
enclose the additional data.
{0102)
Next, Step 1420 for selecting the Gaussian basis
15 functions on the basis of the distance neighborhood
compares the distance neighborhood of each Gaussian
basis function obtained at Step 1410 with a preset
standard value and if it is lower than or equal to the
standard value, it is assumed as a radius adjustment
20 object.
(0103)
The schematic diagram in Fig. 8 shows the case that
the standard value is set to 3 and the Gaussian basis
functions that the distance neighborhood is less than
2 5 or equal to 3 are a radius adjustment object (displayed
- 39 -
as a gray point in the drawing).
{0104)
If the concerned standard value is increased, the
number of the basis functions becoming a radius
5 adjustment object is increased, and finer model
adjustment can be executed, though the calculation cost
is increased. On the other hand, if the concerned
standard value is decreased, the calculation cost can
be inversely decreased, though the model accuracy after
10 correction is comparatively low. The operator of the
plant can optionally set the standard value according
to the needs relating to the accuracy and time.
(01051
Next, Step 1430 adjusts the radial parameters of
15 the Gaussian basis functions selected at Step 1420,
though the detailed adjustment algorithm will be
described later.
{0106}
Next, Step 1440, using the radial parameters of the
20 Gaussian basis functions adjusted, updates the
information saved in the model constructive database
220 and goes to the step for ending the statistical
model adjustment algorithm.
{0107}
Fig. 9 is a flow chart showing the detailed
- 40 -
algorithm operation of the radial parameter adjustment
of the Gaussian basis functions by the model adjusting
part 600 and is equivalent to Step 1430 for adjusting
the radial parameter in the flow chart in Fig. 7.
5 {0108)
The flow chart shown in Fig. 9 executes a
combination of Steps 1431, 1432, 1433, 1434, 1435, and
1436.
{0109}
10 The model adjusting part 600 in the present
invention, firstly, for the model input conditions xt
(t: the repetition frequency of Steps 1432 to 1436)
which is decided at random in the model property space
obtains the contribution degree to the coverage of xt
15 of each Gaussian basis function on the basis of the
distance information to the selected Gaussian basis
functions. And, the model adjusting part 600 corrects
the radial parameter so that the radial parameter
approaches the radial parameter target value calculated
2 0 on the basis of the contribution degree. By repeatedly
executing the series of operations by the fixed number
of times, it executes the radial parameter adjustment.
{OllO)
After the algorithm start of the radial parameter
25 adjustment, Step 1431 for initializing the repetition
- 41 -
frequency t (t = 1) of the radial parameter adjustment
initializes the repetition frequency t (sets t = 1).
(0111)
Next, Step 1432 for determining the model input
5 conditions at random generates input conditions xt at
random in the model property space.
(0112)
Next, Step 1433 for calculating the contribution
degree of each basis function uses the following
10 Formula (2) which is a functional formula included in
the model adjusting part 600 and calculates a
contribution degree pi of each basis function.
{0113}
{Formula 2)
dmin
j=I 'min
(0114)
In Formula (2), dmin indicates a minimum value of
the distance between the selected Gaussian basis
function and xt and di indicates a distance between the
selected Gaussian basis function and xt. By Formula (2),
25 the contribution degree pi is larger in a basis
- 42 -
function at a smaller distance from xt, and as the
distance becomes longer, it decreases exponentially.
Namely, it means that when covering xt, as the basis
function is closer to xt, the contribution degree of
5 the coverage is larger.
(0115)
Fig. 10 is a schematic diagram of the process in
the two-dimensional input space of Steps 1432 and 1433,
and each gray point indicates the central coordinates
10 of the Gaussian basis function which becomes a radius
adjustment object, and each white point indicates the
central coordinates of the Gaussian basis function
which does not become a radius adjustment object, and X
indicates xt. The numerical value on the right shoulder
15 of the selected Gaussian basis function indicates a
contribution degree and as approaching xt, it is found
that the value increases.
{0116}
Next, Step 1434 for calculating the coverage target
2 0 value of each basis function, on the basis of the
calculated contribution degree, calculates a coverage
target value vik of each basis function at xt using the
following Formula (3) which is a functional formula
included in the model adjusting part 600. Here, the
25 coverage is defined by the total of the Gaussian basis
- 43 -
functions obtained by substituting certain model input
conditions for Formula (1) .
{0117)
{Formula 3 )
h = *',
{0118)
Here, the coverage target value qik means a
coverage target value relating to the k component of
the model output of the Gaussian basis function i. Ck
10 is a fixed value peculiarly set according to the model
output component and qik is a value when Ck is
proportionally distributed according to the contribute
degree. Ck is appropriately set, thus the radial
parameter of each basis function can be adjusted so
15 that the coverage under optional model input conditions
becomes a desirable value, and the estimation accuracy
of the statistical mode can be improved.
{0119)
Next, Step 1435 for updating the radial parameter
20 of each basis function, for the calculated coverage
target value, updates a radial parameter rik according
to the following Formula (4) and Formula (5) which are
functional formulas included in the model adjusting
part 600.
{0120}
{Formula 4)
(0121)
{Formula 5)
In Formula (4), r*ik is an update target value of
10 the radial parameter rik and it is decided so that the
basis functional value at xt of the Gaussian function
basis i coincides with cpik. Further, in Formula (5), a
is a step size parameter (< 1) of the radius adjustment
and rik is updated so as to approach the radius update
15 target value r*ik.
{0123}
The next Step 1438 for determining whether t has
reached the maximum repetition frequency or not is a
branch. If the repetition frequency t becomes larger
2 0 than or equal to the preset maximum repetition
frequency, the process goes to the step for ending the
radial parameter adjustment algorithm and in other
cases, t = t + 1 is set, and the process is returned to
Step 1432.
{0124}
As is obvious from the above series of explanation,
the model adjusting part 600 of the control device 200,
for the model construction data added by preprocessing
the plant measurement data, adjusts only the radial
5 parameter of the Gaussian basis function in the
neighborhood of the addition data, thereby can reduce
the calculation cost.
Furthermore, by executing the learning in
10 combination of the three kinds of model construction
data based on the data confidence interval, various
control according to the control needs of the plant is
enabled. This ends the explanation of the detailed
operation of the model adjusting part 600.
Next, in the plant control device which is this
embodiment, the screen displayed on the display device
920 for displaying the maintenance tool output signal
94 transmitted from the outside output interface 913 of
2 0 the maintenance tool 910 capable of transmitting and
receiving data to and from the control device 200 will
be explained by referring to Figs. 11, 12, and 13. Figs
11 to 13 show one concrete example of the screen
displayed on the display device 920.
(0127)
- 46 -
Fig. 11, in the plant control device which is this
embodiment, shows a screen example displayed on the
display device 920 when setting the execution
conditions of the data pre-processing part 300, the
5 model adjusting part 600, and the operation learning
part 800 which are included in the control device 200
and shows an example of the screen specification of
Step 1000 for setting the execution conditions in the
flow chart of Fig. 2 showing the operation procedure of
10 the plant control device of this embodiment.
(0128)
On the setting screen of the execution conditions
shown in Fig. 11, the radius adjustment algorithm
parameters of the statistical models used by the model
15 adjusting part 600, the model input bias and the
confidence interval which are used by the data preprocessing
part 300, the model output target conditions
at the time of learning used by the operation learning
part 800, and the selection of the statistical models
2 0 to be learned can be respectively set.
(0129)
In the state that the screen shown in Fig. 11 is
displayed on the display device 920, the mouse 902 of
the external data input equipment 900 is operated, and
2 5 the focus is moved to the numerical box on the screen,
- 47 -
and numerical values can be input using the keyboard
901. Further, the mouse 902 is operated and the button
on the screen is clicked, thus the button can be
selected (pressed). Similarly, the mouse 902 is
5 operated and the checkbox on the screen is clicked,
thus the checkbox can be checked.
(0130)
On the screen shown in Fig. 11, firstly, the
parameters (Ck, a) used for the radius adjustment
10 algorithm of the statistical model in the model
adjusting part 600 are set. Concretely, in a numerical
box 3001 for the parameter displayed in a parameter
item list 3000, a set value is input.
{0131)
Next, the model input bias (Ax) and model output
confidence interval set value ( ) which are used by the
statistical process in the data pre-processing part 300
are set from numerical boxes 3002 and 3003 respectively,
which are shown in Fig. 11.
{0132)
Furthermore, in the setting list of the target
conditions shown in Fig. 11, the target conditions of
the model output value considered at the time of
learning of the operation method by the operation
2 5 learning part 800 are set. Concretely, in a numerical
- 48 -
box 3005 for each model output displayed in a model
output list 3004, the set value of the target condition
is input.
{0133)
5 Finally, in the setting list of the learning object
models shown in Fig. 11, as for the statistical models
constructed using the model construction data prepared
by the data pre-processing part 300, the statistical
model which is a learning object of the operation
10 method is selected. Concretely, for each model output
displayed in a model output list 3006, checkboxes 3007,
3008, 3009, and 3010 of each item of "Maximum",
"Average", "Minimum", and "All" are respectively
selected, thus the statistical models which are a
15 learning object of the operation learning part 800 can
be selected. The operation learning part 800, for each
of the selected statistical models, learns the
operation method.
(0134 1
After the end of the aforementioned execution
condition setting, if a button 3011 is selected, the
execution condition setting screen ends and the process
goes to the execution of Step 1100 in Fig. 2.
{0135)
Fig. 12, in the plant control device which is this
- 49 -
embodiment, when displaying the learning results for
the statistical model 500 of a learning object which
are set on the execution condition setting screen shown
in Fig. 11, shows a screen example displayed on the
5 display device 920.
(0136)
In the state that the screen shown in Fig. 12 is
displayed on the display device 920, the mouse 902 of
the external data input equipment 900 is operated and
10 the button on the screen is clicked, thus the button
can be selected (pressed).
I01371
On the screen shown in Fig. 12, the learning
results for each statistical model 500 before operation
15 and after learning are displayed on the learning result
list. Concretely, for an item ID column 3100, the
forecasting values of a model input value 3101 and a
model output value 3102 before operation and after
operation using each learning result and the
20 statistical model setting 3103 selected at the time of
learning are displayed.
I01381
In the concerned list, the learning results for the
combination of all the statistical models set on the
2 5 image display in Fig. 11 are displayed and it can be
- 50 -
discriminated by the statistical model setting 3103.
The plant operator, among the learning results
displayed in the list, can select an optional learning
result in consideration of the control needs. The
5 selected result is highlighted by an offset 3104.
Thereafter, if a button 3105 is selected, the screen
shown in Fig. 13 is started.
(0139)
Fig. 13, in the plant control device which is this
10 embodiment, shows the model property of the statistical
model 500 for the learning result selected on the
learning result list display screen and a model inputoutput
condition value 3203 before operation.
(0140)
15 Fig. 13, furthermore to confirm a forecasting value
3204 of the model input-output condition after
operation based on the learning results, is a screen
displayed on the display device 920 and is an example
of the screen used at the time of control execution
2 0 decision of Step 1900 in the control operation flow of
the plant control device of this embodiment shown in
Fig. 2.
{0141)
In the state that the screen shown in Fig. 13 is
2 5 displayed on the display device 920, the mouse 902 of
- 51 -
the external data input equipment 900 is operated, and
the focus is moved to the numerical box on the screen,
and numerical values can be input using the keyboard
901. Further, the mouse 902 is operated and the button
5 on the screen is clicked, thus the button can be
selected (pressed). Similarly, the mouse 902 is
operated and the checkbox on the screen is clicked,
thus the checkbox can be checked.
(01421
10 On the screen shown in Fig. 13, in a graph area
3201, the property graph of the statistical model
calculated by the statistical model 500 corresponding
to the selected learning result is shown. At that time,
to express the statistical model property composed of
15 multi-input and multi-output on a graph as simply as
possible, the screen displays the statistical model
property which is respectively mapped to the 1 input
and 1 output properties.
to1431
As shown on the screen in Fig. 13, in the order
from the upper left of the graph area 3201, the model
input 1-model output 1 property, the model input 1-
model output 2 property, ..., are successively displayed.
In each graph, a property curve 3202 of the model
2 5 output for the one-dimensional model input space is
- 52 -
shown. Further, the model input-output conditions
before operation and the optimum model input-output
conditions which are shown in Fig. 12 after learning
are stacked and displayed on the graph as a plot point
5 3203 and a plot point 3204. By doing this, the plant
operator, on the model property graph, can confirm that
the plant operation based on the learning results is
appropriate.
(01441
10 On the screen shown in Fig. 13, the checkbox 3207
is selected furthermore, thus the Gaussian basis
function used for construction of the statistical model
can be displayed as a figure 3206 in the graph. By
doing this, how the radial parameter of the Gaussian
15 basis function has been adjusted by the radius
adjustment algorithm in the control device of this
embodiment can be confirmed.
{0145}
Furthermore, a checkbox 3208 on the screen shown in
20 Fig. 13 is selected, thus the target condition of each
model output set by screen display of Fig. 11 can be
displayed in the graph as a dotted line 3205. By doing
this, whether the operation based on the learning
results satisfies appropriately the target conditions
2 5 or not can be determined.
(01461
The plant operator confirms the screen display
contents having the aforementioned characteristics, and
then can determine whether to execute the operation.
5 When executing the operation since the learning results
are determined to be appropriate, a button 3209 on the
screen shown in Fig. 13 is selected and in other cases,
a button 3210 is selected.
{0147}
10 The above are the explanations of the screen
displayed on the display device 920 of the plant
control device of this embodiment.
{0148}
As explained above, according to this embodiment,
15 when using the measurement data of the plant and
adjusting the statistical model on line by the RBF
network, a control device for a plant that can end the
optimization calculation within the control cycle and
improves the estimation accuracy of the statistical
2 0 model can be implemented.
{Embodiment 21
(0149)
Next, a control device for a thermal power plant
which is a second embodiment that the plant control
25 device relating to the present invention is applied to
- 54 -
the thermal power plant will be explained by referring
to Figs. 14 and 15.
(0150)
The constitution and function of the plant control
5 device 200 applied to the control device for the
thermal power plant of this embodiment are the same as
those of the control device 200 of the plant control
device of the first embodiment shown in Figs. 1 to 13,
so that to explain here the control device 200 will be
10 omitted.
Further, even when the plant control device
relating to the present invention controls a plant
except the thermal power plant, it is needless to say
that the plant control device 200 of the first
15 embodiment shown in Figs. 1 to 13 can be applied.
(0151)
Fig. 14 shows the control device for the thermal
power plant of this embodiment and is a schematic
diagram showing the constitution of a thermal power
20 plant 100a, to which the control device 200 relating to
this embodiment is applied. Firstly, the power
generation mechanism by the thermal power plant lOOa
will be briefly explained.
I01521
In Fig. 14, in the boiler 101 composing the thermal
- 55 -
power plant 100a, a plurality of burners 102 supplied
with pulverized coal, which is fuel produced by
pulverizing coal finely by the mill 110, and primary
air for transferring pulverized coal and secondary air
5 for combustion adjustment are installed and the
pulverized coal supplied via the burners 102 is burned
inside the boiler 101. Further, the pulverized coal and
the primary air are led from a pipe 134 to the burners
102 and the secondary air is led from a pipe 141 to the
10 burners 102.
(0153)
Further, in the boiler 101, an after air port 103
for introducing double-combustion air into the boiler
101 is installed. The double-combustion air is led from
15 a pipe 142 to the after air port 103.
I01541
High-temperature combustion gas generated by
burning pulverized coal inside the boiler 101 flows
down on the downstream side along the path inside the
2 0 boiler 101, exchanges heat with feed water by a heat
exchanger 106 arranged inside the boiler 101, generates
steam, then becomes exhaust gas, flows into an air
heater 104 installed on the downstream side of the
boiler 101, exchanges heat with the air heater 104, and
25 raises the air fed to the boiler 101 in temperature.
- 56 -
(0155)
And, the exhaust gas passing through the air heater
104 is performed with an exhaust gas process, which is
not shown, and is discharged into the atmosphere from
5 the chimney.
(01561
The feed water circulating the heat exchanger 106
of the boiler 101 is fed to the heat exchanger 106 via
a water feed pump 105, is overheated by the combustion
10 gas flowing down through the boiler 101 in the heat
exchanger 106, and becomes high-temperature and highpressure
steam. Further, in this embodiment, the number
of heat exchangers is set to 1, though a plurality of
heat exchangers may be arranged.
15 (0157)
The high-temperature and high-pressure steam
generated by the heat exchanger 106 is led to a steam
turbine 108 via a turbine governor 107, drives the
steam turbine 108 by the energy possessed by the steam,
20 and generates electric power by a generator 109.
(0158)
In the aforementioned thermal power plant 100a of
this embodiment, various measuring instruments for
detecting the state quantities showing the operation
2 5 state of the thermal power plant are arranged.
(01591
The thermal power plant lOOa is concerned to the
plant 100 shown in Fig. 1, so that the measurement
signal data of the thermal power plant acquired from
5 these measuring instruments, similarly to the plant
control device 200 shown in Fig. 1, is transmitted to
the outside input interface 201 of the control device
200 as the measurement signal data 1 from the plant 100
{0160)
10 As the measuring instruments, for example, as shown
in the thermal power plant lOOa in Fig. 14, a
temperature measuring instrument 151 for measuring the
temperature of the high-temperature and high-pressure
steam fed from the heat exchanger 106 to the steam
15 turbine 108, a pressure measuring instrument 152 for
measuring the steam pressure, and a power generation
output measuring instrument 153 for measuring the
electric power quantity generated by the generator 109
are shown.
2 0 {0161}
The feed water generated by cooling steam by a
condenser (not shown) of the steam turbine 108 is fed
to the heat exchanger 106 of the boiler 101 by the
water feed pump 105, though the feed water flow rate is
2 5 measured by a flow rate measuring instrument 150.
{0162}
Further, the measurement signal data of the state
quantities relating to the concentrations of the
components (such as nitrogen oxide (NOx), carbon
5 monoxide (CO), and hydrogen sulfide (H2S) ) included in
the exhaust gas which is combustion gas discharged from
the boiler 101 is measured by a concentration measuring
instrument 154 installed on the downstream side of the
boiler 101.
10 (01631
Namely, in the control device of the thermal power
plant of this embodiment applied to the thermal power
plant 100a, in the measurement data items of the
thermal power plant 100a which are measured by the
15 measuring instruments 150 to 154 and are input to the
control device 200, the fuel flow rate fed to the
boiler 101 which is the state quantity of the thermal
power plant lOOa measured by each measuring instrument
described above, the air flow rate fed to the boiler
20 101, the feed water flow rate fed to the heat exchanger
106 of the boiler 101, the steam temperature generated
by the heat exchanger 106 of the boiler 101 and fed to
the steam turbine 108, the water feed pressure of the
feed water fed to the heat exchanger 106 of the boiler
2 5 101, the gas temperature of the exhaust gas discharged
- 59 -
from the boiler 101, the gas concentration of the
exhaust gas, and the exhaust gas recirculation flow
rate when a part of the exhaust gas discharged from the
boiler 101 is re-circulated to the boiler 101 are
included.
{ 0164 1
These measurement data items are the measurement
data items decided by the control signal 15 calculated
and output by the control signal generation part 700 in
the plant control device 200 shown in Fig. 1.
I01651
Further, generally, other than the measuring
instruments shown in Fig. 14, many measuring
instruments are arranged in the thermal power plant
100a, though here, the illustration of them is omitted.
I01661
Next, the paths of air input into the boiler 101,
that is, the paths of the primary air and the secondary
air which are input into the boiler 101 from the
burners 102 and the path of air input from the after
air port 103 into the boiler 101 will be explained
using the thermal power plant 100a shown in Fig. 14.
{0167)
In the boiler 101 of the thermal power plant lOOa
shown in Fig. 14, the primary air is led from a fan 120
into a pipe 130 and it branches on the way to a pipe
132 passing through the air heater 104 installed on the
downstream side of the boiler 101 and a pipe 131
bypassing without passing through the air heater 104,
5 though it becomes a pipe 133 arranged on the downstream
side of the air heater 104 and joins again, and is led
to the mill 110 for producing pulverized coal which is
installed on the upstream side of the burners 102.
(0168)
10 The primary air passing through the air heater 104
is heated by exchanging heat with combustion gas
flowing down the boiler 101. The primary air bypassing
the air heater 104 together with the heated primary air
transmits the pulverized coal milled by the mill 110 to
15 the burners 102.
{0169)
The air input from a pipe 140 using a fan 121 is
heated similarly by the air heater 104 and then
branches to the pipe 141 for the secondary air and the
2 0 pipe 142 for the after air port and they are led
respectively to the burners 102 of the boiler 101 and
the after air port 103.
(0170)
In the control device 200 for the thermal power
2 5 plant of this embodiment, as an example of controlling
- 61 -
the air flow rate transmitted from the fan 121 and
input from the burners 102 and the after air port 103
into the boiler 101, an air damper 162 and an air
damper 163 are installed as operation end devices on
5 the upstream side of the pipe 141 for the secondary air
and the pipe 142 for the after air port are installed
and structured so that the openings of the air damper
162 and the air damper 163 can be adjusted by the
control device 200 to allow controlling the flow rate
10 of the secondary air fed into the boiler 101 and the
flow rate of the after-air.
(0171)
Further, as an example of controlling the air flow
rate transmitted from the fan 120 and input into the
15 boiler 101 together with pulverized coal from the
burners 102, in the pipe 131 and the pipe 132 which are
just prior portions joining to the pipe 133, an air
damper 160 and an air damper 161 are installed as
operation end devices and structured so that the
20 openings of the air damper 160 and the air damper 161
can be adjusted by the control device 200 to allow
controlling the flow rate of the air fed into the
boiler 101.
(01721
25 The control device 200 can control other
- 62 -
measurement data items, so that the installation place
of the operation end device may be changed according to
the control object.
I01731
5 Fig. 15 is an enlarged drawing of the pipe part
relating to the air heater 104 installed on the
downstream side of the boiler 101 of the thermal power
plant 100a shown in Fig. 14.
I01741
10 As shown in Fig. 15, in the air heater 104, the
pipe 130 and the pipe 140 for feeding air are installed,
and among them, the pipe 140 is arranged passing
through the air heater 104, and the pipe 130 is
composed of the pipe 131 and the pipe 132 which are
15 branched on the way, and the pipe 131 is arranged by
bypassing the air heater 104, and the pipe 132 is
arranged passing through the air heater 104.
(0175)
And, the pipe 132 is arranged so as to pass through
2 0 the air heater 104, then become the pipe 133 joining to
the pipe 131, be led to the mill 110, and lead air
together with pulverized coal to the burners 102 of the
boiler 101 via the concerned pipe 133 from the mill 110.
I01761
Further, the pipe 140 passes through the air heater
- 63 -
104 and then branches to the pipe 141 and the pipe 142,
and among them, the pipe 141 and the pipe 142 are
arranged so as to lead air to the burners 102 of the
boiler 101 and the after air port 103 of the boiler 101,
respectively.
to1771
Further, on the pipe 131 and the pipe 132 of the
just prior portion joining to the pipe 133, the air
damper 160 and the air damper 161 for adjusting the
ventilating air quantity are installed respectively and
on the upstream side of the pipe 141 and the pipe 142,
the air damper 162 and the air damper 163 for adjusting
the ventilating air quantity are installed respectively.
{0178}
And, by operating these the air dampers 160 to 163,
the air passing area can be changed by the pipes 131,
132, 141, and 142, so that the air flow rates passing
through the pipes 131, 132, 141, and 142 and fed into
the boiler 101 can be adjusted individually.
I01791
The control signal 15 calculated by the control
signal generation part 700 of the control device 200 is
output as the operation signal 16 for the thermal power
plant lOOa via the outside output interface 202 and the
control end devices such as the air dampers 160, 161,
- 64 -
162, and 163 respectively installed in the pipes 131,
132, 141, and 142 of the boiler 101 are operated.
{0180}
Further, in this embodiment, the devices such as
5 the air dampers 160, 161, 162, and 163 are called the
operation ends and the output signal instructed to the
operation end from the control device 200 by the
control signal 15 calculated by the control device 200
necessary to operate them is called the operation
10 signal 16.
{Ol8l}
Further, the operation signal 16 which is
calculated by the control signal generation part 700
and is output to the operation end includes the air
15 flow rate fed to the boiler 101 via the pipes 131, 132,
141, and 142, the openings of the air dampers 160 to
163 for adjusting the air flow rate respectively
installed in the pipes 131, 132, 141, and 142 for
feeding air to the boiler 101, the fuel flow rate of
2 0 pulverized coal supplied to the burners 102 of the
boiler 101, and the exhaust gas recirculation flow rate
for recirculating a part of the exhaust gas discharged
from the boiler 101 to the boiler 101.
I01821
25 Hereinafter, in the control device for the thermal
- 65 -
power plant of this embodiment, the case that the
operation ends installed in the thermal power plant
lOOa controlled by the control device 200 are assumed
as the air dampers 160 and 161 respectively installed
in the pipes 131 and 132 for adjusting the air quantity
fed to the burners 102 installed in the boiler 101 and
as the air dampers 162 and 163 respectively installed
in the pipes 141 and 142 for adjusting the air quantity
fed to the after air port 103 installed in the boiler
101, and the controlled quantity is assumed as the
concentrations of CO, NOx, and H2S in the exhaust gas
discharged from the boiler 101 will be explained.
I01831
Further, in the control device of the thermal power
plant of this embodiment, the operation quantity (the
openings of the air dampers 160, 161, 162, and 163) of
the operation ends of the boiler 101 becomes the model
input of the statistical model 500 composing the
control device 200, and the concentrations of NOx, CO,
and H2S included in the exhaust gas discharged from the
boiler 101 become model output of the statistical model
500, and the respective miniaturization of the model
input and output is a learning object.
I01841
As explained above, if the plant control device of
- 66 -
the present invention is applied to the thermal power
plant, by learning the operation method satisfying the
demands for the environmental restrictions and
operation cost, the target values of the concentrations
of NOx, CO, and H2S discharged from the thermal power
plant can be accomplished.
{0185}
According to this embodiment, a control device for
a thermal power plant having the functions of using the
plant measurement data and adjusting the statistical
model by the RBF network online, suppressing the
influence on the calculation cost due to accumulation
of the measurement data, adjusting the statistical
model within the control cycle, and improving the
estimation accuracy can be implemented.
{0186}
Further, the present invention is not limited to
the aforementioned embodiments and includes various
modifications. For example, the aforementioned
embodiments are explained in detail to understandably
explain the present invention and are not always
limited to the embodiments including all the explained
constitutions. Further, a part of the constitution of a
certain embodiment can be replaced with the
constitution of other embodiments and to the
- 6 7 - '
constitution of a certain embodiment, the constitution
of other embodiments can be added. Further, for a part
of the constitution of each embodiment, other
constitutions can be added, deleted, and replaced.
(0187)
Further, each constitution, function, processing
part, and processing means described above may be
implemented by hardware, for example, by designing a
part of them or the whole by an integrated circuit.
Further, each constitution and function described above
may be implemented by software by interpreting and
executing programs for implementing the respective
functions by the processor. A program, a table, a file,
measurement information, and information such as
calculation information for implementing each function
can be stored in the archive device such as a memory, a
hard disk, and an S S D (solid state drive) or the
archive medium such as an IC card, an S D card, and a
DVD. Therefore, each process and each constitution, as
a processing part, a processing unit, and a program
module, can implement each function.
I01881
Further, the control line and information line
which are considered to be necessary from the viewpoint
of explanation are shown and all the control lines and
- 68 -
information lines are not always shown from the
viewpoint of products. Actually, almost all the
constitutions may be considered to be connected
mutually.
(01891
As explained above, according to this embodiment,
when using the measurement data of the plant and
adjusting the statistical model online by the RBF
network using the plant measurement data, a control
device for a thermal power plant that can end the
optimization calculation within the control cycle and
that improves the estimation accuracy of the
statistical model can be implemented.
{Industrial Applicability)
(0190)
The present invention can be applied to a control
device for a plant and a control device for a thermal
power plant.
{Reference Signs List)
(0191)
1: Measurement signal data, 16: Control signal, 90:
Input-output data information, 100: Plant, 100a:
Thermal power plant, 101: Boiler, 102: Burner, 103:
After air port, 130 to 133: Pipe, 140 to 142: Pipe, 160
to 163: Air damper, 200: Control device, 201: Outside
- 69 -
input interface, 202: Outside output interface, 210:
Measurement signal database, 220: Model constructive
database, 230:Learning information database, 240:
Control logic database, 250: Control signal database,
5 300: Data pre-processing part, 400: Numerical
calculation part, 500: Statistical model, 600: Model
adjusting part, 700: Control signal generation part,
800: Operation learning part, 900: External data input
equipment, 901: Keyboard, 902: Mouse, 910: Maintenance
10 tool, 911: Outside input interface, 912: Data
transmitting and receiving part, 913: Outside output
interface, 920: Display device

WE CLAIM:
{Claim 1)
A control device for a plant for taking in
measurement signal data, which is a state quantity of
5 the plant, from the plant and calculating an operation
signal for controlling the plant using the measurement
signal data, characterized in that,
the control device is comprising a measurement
signal database for taking in and saving the
10 measurement signal data which is the state quantity of
the plant; a data pre-processing part for executing a
statistical process on the basis of a confidence
interval for the measurement data of the plant which
has been saved in the measurement signal database,
15 thereby generating model construction data; a model
constructive database for saving the model construction
data generated by the data pre-processing part; a
statistical model for simulating a control property of
the plant for estimating a value of the measurement
2 0 signal data which is the state quantity of the plant
when a control signal is given to the plant using the
model construction data saved in the model constructive
database; an operation learning part for learning a
generation method of model input equivalent to the
25 control signal given to the plant so that model output
Amended sheet (Article 19)
- 71 -
equivalent to the measurement signal data accomplishes
its target value using the statistical model; a
learning information database for saving learning data
relating to learning restrictions and learning results
5 in the operation learning part; and a control signal
generation part for calculating the control signal
transmitted to the plant using the measurement signal
data of the measurement signal database and the
learning data of the learning information database,
10 wherein the control device is further comprising a
model adjusting part for adjusting a basis radial
parameter of the statistical model included in the
model construction data saved in the model constructive
database so that the statistical model generates the
15 model output using adjustment results of the basis
radial parameter by the model adjusting part and the
model output is output to the operation learning part.
{Claim 2)
The control device for a plant according to Claim 1,
2 0 the data pre-processing part has a function, using
the measurement signal data saved in the measurement
signal database, of preparing at least one piece of
data among the model construction data classified in
three kinds of maximum, average, and minimum in
25 consideration of the confidence interval of the data.
Amended sheet (Article 19)
- 72 -
{Claim 3 )
The control device for a plant according to Claim 1,
information saved in the model constructive
database includes at least one piece of information
5 among a model input value of each piece of data, model
output values of three kinds of maximum, average, and
minimum for it, a radial parameter value of a Gaussian
basis function arranged on each piece of data, and a
weight parameter value of a RBF network composing the
10 statistical model constructed using the data.
{Claim 4 )
The control device for a plant according to Claim 1,
a RBF network composing the statistical model
constructed using the model construction data saved in
15 the model constructive database has a structure that in
each node of an intermediate layer, (model output
component number) x 3 Gaussian function basis are
included.
{Claim 5)
2 0 The control device for a plant according to Claim 1,
the model adjusting part has a function that uses
an index of a distance neighborhood when determining a
Gaussian basis function of an adjustment object of the
radial parameter;
2 5 in a calculation of the distance neighborhood, for
Amended sheet (Article 19)
- 73 -
the Gaussian basis functions grouped on the basis of
the measurement data obtained from the plant,
determines the distance neighborhood of the basis
function where a distance from the measurement data is
5 minimized in each group as 1; and
for other basis functions, calculates the distance
neighborhood so as to increase the distance
neighborhood by 1 each in the ascending order of the
distance from the measurement data.
10 {Claim 6)
(After amendment)
The control device for a plant according to Claim 5,
the model adjusting part adjusts the radial
parameter for a Gaussian basis function where the
15 distance neighborhood is smaller than or equal to a
preset standard value.
{Claim 7)
The control device for a plant according to Claim 1,
the model adjusting part has a function, in the
2 0 adjustment of the radial parameter, for considering a
contribution degree of each Gaussian basis function
under a model input condition decided at random and in
a calculation of the contribution degree, for
calculating the contribution degree of the Gaussian
25 basis function where a distance from the model input
Amended sheet (Article 19)
- 74 -
condition is minimized as 1, and in other basis
functions, for calculating the contribution degree so
as to decrease the contribution degree exponentially
according to the distance from the model input
5 condition.
{Claim 8)
(After amendment)
The control device for a plant according to Claim 7,
the model adjusting part has a function that, in
10 the radial parameter adjustment, for a model input
condition determined at random, calculates the radial
parameter target value so that a coverage target value
predetermined by each Gaussian basis function coincides
with a value proportionally allotted by the
15 contribution degree and that updates the radial
parameter so that the radial parameter approaches the
radial parameter target value.
{Claim 9)
The control device for a plant according to Claim 1,
2 0 the operation learning part has a function, for a
statistical model optionally selected from a plurality
of statistical models constructed using the model
construction data adjusted by the model adjusting part,
for learning an optimum model input generation method.
25 {Claim 10)
Amended sheet (Article 19)
- 75 -
The control device for a plant according to Claim 1,
the control device is connected to a display device
and includes an output part for displaying a list of
results learned by the operation learning part on the
5 display device for the plurality of statistical models
constructed using the model construction data.
{Claim 11)
A control device for a thermal power plant for
taking in measurement signal data, which is a state
10 quantity of the plant, from the thermal power plant
including a boiler and calculating an operation signal
for controlling the plant using the measurement signal
data, characterized in that,
the control device is comprising a measurement
15 signal database for taking in and saving the
measurement signal data which is a state quantity of
the plant; a data pre-processing part for executing a
statistical process on the basis of a confidence
interval for the measurement data of the plant which
2 0 has been saved in the measurement signal database,
thereby generating model construction data; a model
constructive database for saving the model construction
data generated; a statistical model for simulating a
control property of the plant for estimating a
2 5 measurement signal data value which is the state
Amended sheet (Article 19)
- 76 -
quantity of the plant when a control signal is given to
the plant using the model construction data saved in
the model constructive database; an operation learning
part for learning a generation method of model input
5 equivalent to the control signal given to the plant so
that model output equivalent to the measurement signal
data accomplishes its target value using the
statistical model; a learning information database for
saving learning data relating to learning restrictions
10 and learning results in the operation learning part;
and a control signal generation part for calculating
the control signal transmitted to the plant using the
measurement signal data of the measurement signal
database and the learning data of the learning
15 information database,
wherein the control device is further comprising a
model adjusting part for adjusting a basis radial
parameter of the statistical model included in the
model construction data saved in the model constructive
2 0 database so that the statistical model generates the
model output using adjustment results of the basis
radial parameter by the model adjusting part and the
model output is output to the operation learning part,
and
2 5 the measurement signal data includes a signal of
Amended sheet (Article 19)
the state quantity expressing at least one of
concentrations of nitrogen oxide, carbon monoxide,
carbon dioxide, nitrogen, oxygen, unburnt carbon in ash,
and hydrogen sulfide, and
5 the control signal includes a signal expressing at
least one of an air flow rate fed to the boiler of the
thermal power plant, an aperture of an air dumper for
adjusting the air flow rate, a fuel flow rate supplied
to the boiler, and an exhaust gas recirculation flow
10 rate for recirculating exhaust gas discharged from the
boiler to the boiler.
{Claim 12)
The control device for a thermal power plant
according to Claim 11,
15 the data pre-processing part has a function, using
the measurement signal data saved in the measurement
signal database, for preparing at least one piece of
data among the model construction data classified in
three kinds of maximum, average, and minimum in
2 0 consideration of the confidence interval of the data.
{Claim 13)
The control device for a thermal power plant
according to Claim 11,
information saved in the model constructive
2 5 database includes at least one piece of information
Amended sheet (Article 19)
- 78 -
among a model input value of each piece of data, model
output values of three kinds of maximum, average, and
minimum for it, a radial parameter value of a Gaussian
basis function arranged on each piece of data, and a
5 weight parameter value of a RBF network composing the
statistical model constructed using the data.
(Claim 14)
The control device for a thermal power plant
according to Claim 11,
10 a RBF network composing the statistical model
constructed using the model construction data saved in
the model constructive database has a structure that in
each node of an intermediate layer, (model output
component count) x 3 Gaussian function basis are
15 included.
{Claim 15)
The control device for a thermal power plant
according to Claim 11,
the model adjusting part has a function that
2 0 uses an index of a distance neighborhood when
determining a Gaussian basis function of an adjustment
object of the radial parameter;
in a calculation of the distance neighborhood, for
the Gaussian basis functions grouped on the basis of
2 5 the measurement data obtained from the plant,
Amended sheet (Article 19)
- 79 -
determines the distance neighborhood of the basis
function where a distance from the measurement data is
minimized in each group as 1; and
for other basis functions, calculates the distance
5 neighborhood so as to increase the distance
neighborhood by 1 each in the ascending order of the
distance from the measurement data.
{Claim 16)
(After amendment)
10 The control device for a thermal power plant
according to Claim 15,
the model adjusting part adjusts the radial
parameter for a Gaussian basis function where the
distance neighborhood is smaller than or equal to a
15 preset standard value.
{Claim 17)
The control device for a thermal power plant
according to Claim 11,
the model adjusting part has a function, in the
2 0 adjustment of the radial parameter, for considering a
contribution degree of each Gaussian basis function
under a model input condition decided at random and in
a calculation of the contribution degree, for
calculating the contribution degree of the Gaussian
25 basis function where a distance from the model input
Amended sheet (Article 19)
- 80 -
condition is minimized as 1, and in other basis
functions, for calculating the contribution degree so
as to decrease exponentially according to the distance
from the model input condition.
5 {Claim 18)
(After amendment)
The control device for a thermal power plant
according to Claim 17,
the model adjusting part has a function that, in
10 the radial parameter adjustment, for a model input
condition determined at random, calculates the radial
parameter target value so that a coverage target value
predetermined by each Gaussian basis function coincides
with a value proportionally allotted by the
15 contribution degree and that updates the radial
parameter so that the radial parameter approaches the
radial parameter target value.
{Claim 19)
The control device for a thermal power plant
2 0 according to Claim 11,
the operation learning part has a function, for a
statistical model optionally selected from a plurality
of statistical models constructed using the model
construction data adjusted by the model adjusting part,
25 for learning an optimum model input generation method.
&ended sheet (Article 19)
- 81 -
{Claim 2 0 )
The c o n t r o l d e v i c e f o r a thermal power p l a n t
a c c o r d i n g t o Claim 11,
t h e c o n t r o l device i s connected t o a d i s p l a y d e v i c e
5 and i n c l u d e s an o u t p u t p a r t f o r d i s p l a y i n g a l i s t of
r e s u l t s l e a r n e d by t h e o p e r a t i o n l e a r n i n g p a r t on t h e
d i s p l a y d e v i c e f o r t h e p l u r a l i t y of s t a t i s t i c a l models
c o n s t r u c t e d using t h e model c o n s t r u c t i o n d a t a .
10 Dated this lSt day of July 2013
Of Anand and Anand Advocates
Agent for the Applicant
Amended sheet ( A r t i c l e 19)